Kimi 开源 FlashKDA 注意力内核
We've open-sourced FlashKDA, our high-performance CUTLASS-based implementation of Kimi Delta Attention kernels.
AI 导读
我们开源了 FlashKDA，这是我们基于 CUTLASS 的 Kimi Delta Attention 内核高性能实现。 在 H20 上，它相比 flash-linear-attention 基线带来 1.72×–2.22× 的 prefill 加速，并可作为 flash-linear-attention 的即插即用后端。
正文
当前显示原文，中文全文译文尚未就绪。
We've open-sourced FlashKDA, our high-performance CUTLASS-based implementation of Kimi Delta Attention kernels.
It delivers 1.72×–2.22× prefill speedup over the flash-linear-attention baseline on H20, and works as a drop-in backend for flash-linear-attention.